98
JUST IN: Advanced AI Models Refuse Military Queries at Alarming Rates, New Report Finds
(www.nationaldefensemagazine.org)
Yeah Fuck AI!!
From words and phrases like “kill chain,” “assault the objective,” “warfighter” and “moving of ammo” to questions about weapon systems, the models don’t “like any of it,” Saltsman said. “They're so overly sensitive that they just won't be helpful.”
Oh what the god damn hell am I in a fucking satire!?
The fact that Elon Musk tried to neuter Grok a million times already, and Grok will still reply directly to Musk offending him and calling his behavior despicable shows that indeed, if you train AI with the goal of sounding reasonable and logical, it becomes fundamentally opposed to certain world views and actions.
So an AI that would work as a US military advisor, happily targeting the next civilian school with a bomb, would simultaneously be too dumb and ineffective to be able to complete the task.
It isn't "programmed to be reasonable", it parrots the most common internet data
Parent comment didn't say "programmed" they said "trained". Which is correct.
I think what this actually shows is that humans are, on average, good. It makes training an evil chatbot difficult.
unfortunately they will never say radical lefty talking points either because of that
Yeah, we're not going to see LeninGPT either. I suppose "good" might be assigning too much moral weight to being peaceful and nice. It's not "good" when people settle for an unjust peace to avoid conflict simply because it's mean to fight back.
Obviously.